👨💻 Senior Software Developer
📹Programming Streamer: http://twitch.tv/cmgriffing
📽️ YouTuber: http://youtube.com/@cmgriffing
📚Douglas Adams Fanatic
Sans Bullshit Sans is undefeated

@cmgriffing.bsky.social
👨💻 Senior Software Developer
📹Programming Streamer: http://twitch.tv/cmgriffing
📽️ YouTuber: http://youtube.com/@cmgriffing
📚Douglas Adams Fanatic
Sans Bullshit Sans is undefeated
Finishing the pi-autoresearch workshop https://x.com/i/broadcasts/1PKqrNXnQdYGb
You know the meme about people at Google only get promoted for spinning up new products? I think Microsoft has a similar thing but with rebranding existing products.
I counted 10 rats in Redmond on the walk home at 1am
Before you hand an AI agent your merge conflicts, check how it scores on Git.
@cmgriffing.bsky.social's 18-minute tour of GitBench: model comparisons, reasoning levels, token usage. The repo is public, so you can run it against your own models.
I had a reset about to expire so I turned on Ultra mode and set it to Fast just so I could see how the 1% live. Even with Sol 6.1's efficiencies, 5 hour limit used up in 10 minutes but it was like the zoomy effect from Limitless. Worth it
TIL that the gemini plan includes Opus 5.5 access I can use the new Opus without giving any money to Anthropic. I'd be living the dream, Jen!
Link to video: https://www.youtube.com/watch?v=y7Oug2rNB-s
Check out gitbench at https://gitbench.gitkraken.com
Finally put together a video about GitBench
I'm really curious how Cloudflare was getting >500ms median response times from Jev. Maybe making requests from Australia?
CoreWeave Glad I put a max token limit in my request config. Before you point out that the whitespace is part of the input, the input was only 272 tokens. Coreweave padded it as part of the output after that.
Another day, another inference provider (@CoreWeave) padding their output response with a bunch of garbage whitespace characters.
Tubi puts Arachnophobia in the comedy section and I don't find that funny.
I really wasn't expecting to have to create a GitBench v2, yet. But the models are getting really good at git.
@cmgriffing.bsky.social goes whole days without opening an IDE now, mostly thanks to Kepler. He still reaches for one to review and trim what the agent wrote, because go-to-definition beats prompting an agent into minimal code.
His take on whether the IDE is dying:
gitkraken.com/blog/challen...
I'm in this marketing team meeting and I had no Idea they cared so much about ICP. Cool team!
Who do I complain to about GPT Image 1 using my likeness without my permission?
On stream, people think I have a tab hoarding issue. If only they knew about the 186 tabs I have going in Safari on my phone.
Some real world numbers for Jev. I ran Jev in parallel to gpt-oss-120b in a dev instance for my startup. The same ~50k requests went to each one. $0.80 vs $2.09